For many years the most commonly used method for cancer diagnostics is Magnetic Resonance Imaging (MRI). MRI procedure uses non-ionizing radiation to create useful diagnostic images in various imaging sequences. Using these modalities we get images with different characteristics that highlight various features of brain and tumor anatomy. These attributes can affect tumor detection. All these sequences provide us with multidimensional data that might be difficult to check by a human expert. We performed tumor detection with a deep neural network and conducted a statistical...

Using MICCAI BraTS 2018 training datasets [1-3] (multimodal scans of 285 patients with ground truth data) and a deep neural network (U-Net), based on Marcinkiewicz et. al. [4], we performed an analysis of tumor detection.Moreover, we combine three modalities (T1CE, T2, FLAIR) into a multi-modalimage (MMI), creating a new artificial sequence. The topology of the network is presented in Fig. 1. For our experiments, we used an ensemble model consisted of four networks whose results were averaged to produce the final output (data division for...

Regarding the Friedman test, we reject the null hypothesis at a significance level of 0.05 in all network folds. In all cases inequality of at least one median in populations occurs, therefore there is inequal tumor detection at different MRI modalities. According to post-hoc tests, we were able tonotice that similar detectionoccurs in MRI sequences pair T1 - T1 with contrast, in pair FLAIR - MMI (multi-modal image) and in pair T2 - FLAIR. Quantification of the variation of tumor detection using U-Net, describes the...

We were able to show the influence of tumor features on cancer detection. The analysis showed the notable effect of tumor regularity and the influence of tumor size on detection. The larger and more regular the tumor, the better is the detection. Moreover, imbalanced folds with patients with various tumor characteristics (tumor size, tumor regularity) have an association with detection quality thus we state that training data need to be balanced regarding many aspects. Our work allows us to reject the null hypothesis and confirm...

K. Leszczorz; Gliwice/PL - Author at Silesian University of Technology; [email protected] W. Dudzik; Gliwice/PL - Author at Silesian University of Technology; [email protected] J. Nalepa; Gliwice/PL - Author at Silesian University of Technology J. Polanska; Gliwice/PL - Author at Silesian University of Technology This work was financed by the European Union under the European Social Fund (AIDA - POWR.03.02.00-00-I029) and partially financed by the National Science Centre grant no. 2015/19/B/ST6/01736.

B. H. Menze, A. Jakab, S. Bauer, J. Kalpathy-Cramer, K. Farahani, J. Kirby, et al. "The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS)", IEEE Transactions on Medical Imaging 34(10), 1993-2024 (2015) DOI: 10.1109/TMI.2014.2377694 S. Bakas, H. Akbari, A. Sotiras, M. Bilello, M. Rozycki, J.S. Kirby, et al., "Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features", Nature Scientific Data, 4:170117 (2017) DOI: 10.1038/sdata.2017.117 S. Bakas, M. Reyes, A. Jakab, S. Bauer, M. Rempfler, A. Crimi, et al., "Identifying the...